HBASE-21336 Simplify the implementation of WALProcedureMap

This commit is contained in:
zhangduo 2018-10-20 19:33:29 +08:00 committed by Duo Zhang
parent 77ac352a95
commit 3b66b65b9f
9 changed files with 636 additions and 815 deletions

View File

@ -505,8 +505,10 @@ public class ProcedureExecutor<TEnvironment> {
private void loadProcedures(ProcedureIterator procIter, boolean abortOnCorruption)
throws IOException {
// 1. Build the rollback stack
int runnablesCount = 0;
int runnableCount = 0;
int failedCount = 0;
int waitingCount = 0;
int waitingTimeoutCount = 0;
while (procIter.hasNext()) {
boolean finished = procIter.isNextFinished();
@SuppressWarnings("unchecked")
@ -526,11 +528,21 @@ public class ProcedureExecutor<TEnvironment> {
// add the procedure to the map
proc.beforeReplay(getEnvironment());
procedures.put(proc.getProcId(), proc);
if (proc.getState() == ProcedureState.RUNNABLE) {
runnablesCount++;
} else if (proc.getState() == ProcedureState.FAILED) {
failedCount++;
switch (proc.getState()) {
case RUNNABLE:
runnableCount++;
break;
case FAILED:
failedCount++;
break;
case WAITING:
waitingCount++;
break;
case WAITING_TIMEOUT:
waitingTimeoutCount++;
break;
default:
break;
}
}
@ -551,9 +563,10 @@ public class ProcedureExecutor<TEnvironment> {
// have been polled out already, so when loading we can not add the procedure to scheduler first
// and then call acquireLock, since the procedure is still in the queue, and since we will
// remove the queue from runQueue, then no one can poll it out, then there is a dead lock
List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnablesCount);
List<Procedure<TEnvironment>> runnableList = new ArrayList<>(runnableCount);
List<Procedure<TEnvironment>> failedList = new ArrayList<>(failedCount);
Set<Procedure<TEnvironment>> waitingSet = null;
List<Procedure<TEnvironment>> waitingList = new ArrayList<>(waitingCount);
List<Procedure<TEnvironment>> waitingTimeoutList = new ArrayList<>(waitingTimeoutCount);
procIter.reset();
while (procIter.hasNext()) {
if (procIter.isNextFinished()) {
@ -568,15 +581,11 @@ public class ProcedureExecutor<TEnvironment> {
LOG.debug("Loading {}", proc);
Long rootProcId = getRootProcedureId(proc);
if (rootProcId == null) {
// The 'proc' was ready to run but the root procedure was rolledback?
scheduler.addBack(proc);
continue;
}
// The orphan procedures will be passed to handleCorrupted, so add an assert here
assert rootProcId != null;
if (proc.hasParent()) {
Procedure<TEnvironment> parent = procedures.get(proc.getParentProcId());
// corrupted procedures are handled later at step 3
if (parent != null && !proc.isFinished()) {
parent.incChildrenLatch();
}
@ -591,26 +600,10 @@ public class ProcedureExecutor<TEnvironment> {
runnableList.add(proc);
break;
case WAITING:
if (!proc.hasChildren()) {
// Normally, WAITING procedures should be waken by its children.
// But, there is a case that, all the children are successful and before
// they can wake up their parent procedure, the master was killed.
// So, during recovering the procedures from ProcedureWal, its children
// are not loaded because of their SUCCESS state.
// So we need to continue to run this WAITING procedure. But before
// executing, we need to set its state to RUNNABLE, otherwise, a exception
// will throw:
// Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,
// "NOT RUNNABLE! " + procedure.toString());
proc.setState(ProcedureState.RUNNABLE);
runnableList.add(proc);
}
waitingList.add(proc);
break;
case WAITING_TIMEOUT:
if (waitingSet == null) {
waitingSet = new HashSet<>();
}
waitingSet.add(proc);
waitingTimeoutList.add(proc);
break;
case FAILED:
failedList.add(proc);
@ -625,36 +618,31 @@ public class ProcedureExecutor<TEnvironment> {
}
}
// 3. Validate the stacks
int corruptedCount = 0;
Iterator<Map.Entry<Long, RootProcedureState<TEnvironment>>> itStack =
rollbackStack.entrySet().iterator();
while (itStack.hasNext()) {
Map.Entry<Long, RootProcedureState<TEnvironment>> entry = itStack.next();
RootProcedureState<TEnvironment> procStack = entry.getValue();
if (procStack.isValid()) continue;
for (Procedure<TEnvironment> proc : procStack.getSubproceduresStack()) {
LOG.error("Corrupted " + proc);
procedures.remove(proc.getProcId());
runnableList.remove(proc);
if (waitingSet != null) waitingSet.remove(proc);
corruptedCount++;
// 3. Check the waiting procedures to see if some of them can be added to runnable.
waitingList.forEach(proc -> {
if (!proc.hasChildren()) {
// Normally, WAITING procedures should be waken by its children.
// But, there is a case that, all the children are successful and before
// they can wake up their parent procedure, the master was killed.
// So, during recovering the procedures from ProcedureWal, its children
// are not loaded because of their SUCCESS state.
// So we need to continue to run this WAITING procedure. But before
// executing, we need to set its state to RUNNABLE, otherwise, a exception
// will throw:
// Preconditions.checkArgument(procedure.getState() == ProcedureState.RUNNABLE,
// "NOT RUNNABLE! " + procedure.toString());
proc.setState(ProcedureState.RUNNABLE);
runnableList.add(proc);
} else {
proc.afterReplay(getEnvironment());
}
itStack.remove();
}
if (abortOnCorruption && corruptedCount > 0) {
throw new IOException("found " + corruptedCount + " procedures on replay");
}
});
// 4. Push the procedures to the timeout executor
if (waitingSet != null && !waitingSet.isEmpty()) {
for (Procedure<TEnvironment> proc: waitingSet) {
proc.afterReplay(getEnvironment());
timeoutExecutor.add(proc);
}
}
waitingTimeoutList.forEach(proc -> {
proc.afterReplay(getEnvironment());
timeoutExecutor.add(proc);
});
// 5. restore locks
restoreLocks();
// 6. Push the procedure to the scheduler

View File

@ -85,12 +85,18 @@ public interface ProcedureStore {
boolean hasNext();
/**
* Calling this method does not need to convert the protobuf message to the Procedure class, so
* if it returns true we can call {@link #skipNext()} to skip the procedure without
* deserializing. This could increase the performance.
* @return true if the iterator next element is a completed procedure.
*/
boolean isNextFinished();
/**
* Skip the next procedure
* <p/>
* This method is used to skip the deserializing of the procedure to increase performance, as
* when calling next we need to convert the protobuf message to the Procedure class.
*/
void skipNext();

View File

@ -19,7 +19,6 @@ package org.apache.hadoop.hbase.procedure2.store.wal;
import java.io.IOException;
import org.apache.hadoop.fs.FSDataInputStream;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreTracker;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
@ -31,70 +30,25 @@ import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos.ProcedureWALEntry;
/**
* Helper class that loads the procedures stored in a WAL
* Helper class that loads the procedures stored in a WAL.
*/
@InterfaceAudience.Private
public class ProcedureWALFormatReader {
private static final Logger LOG = LoggerFactory.getLogger(ProcedureWALFormatReader.class);
// ==============================================================================================
// We read the WALs in reverse order from the newest to the oldest.
// We have different entry types:
// - INIT: Procedure submitted by the user (also known as 'root procedure')
// - INSERT: Children added to the procedure <parentId>:[<childId>, ...]
// - UPDATE: The specified procedure was updated
// - DELETE: The procedure was removed (finished/rolledback and result TTL expired)
//
// In the WAL we can find multiple times the same procedure as UPDATE or INSERT.
// We read the WAL from top to bottom, so every time we find an entry of the
// same procedure, that will be the "latest" update (Caveat: with multiple threads writing
// the store, this assumption does not hold).
//
// We keep two in-memory maps:
// - localProcedureMap: is the map containing the entries in the WAL we are processing
// - procedureMap: is the map containing all the procedures we found up to the WAL in process.
// localProcedureMap is merged with the procedureMap once we reach the WAL EOF.
//
// Since we are reading the WALs in reverse order (newest to oldest),
// if we find an entry related to a procedure we already have in 'procedureMap' we can discard it.
//
// The WAL is append-only so the last procedure in the WAL is the one that
// was in execution at the time we crashed/closed the server.
// Given that, the procedure replay order can be inferred by the WAL order.
//
// Example:
// WAL-2: [A, B, A, C, D]
// WAL-1: [F, G, A, F, B]
// Replay-Order: [D, C, A, B, F, G]
//
// The "localProcedureMap" keeps a "replayOrder" list. Every time we add the
// record to the map that record is moved to the head of the "replayOrder" list.
// Using the example above:
// WAL-2 localProcedureMap.replayOrder is [D, C, A, B]
// WAL-1 localProcedureMap.replayOrder is [F, G]
//
// Each time we reach the WAL-EOF, the "replayOrder" list is merged/appended in 'procedureMap'
// so using the example above we end up with: [D, C, A, B] + [F, G] as replay order.
//
// Fast Start: INIT/INSERT record and StackIDs
// ---------------------------------------------
// We have two special records, INIT and INSERT, that track the first time
// the procedure was added to the WAL. We can use this information to be able
// to start procedures before reaching the end of the WAL, or before reading all WALs.
// But in some cases, the WAL with that record can be already gone.
// As an alternative, we can use the stackIds on each procedure,
// to identify when a procedure is ready to start.
// If there are gaps in the sum of the stackIds we need to read more WALs.
//
// Example (all procs child of A):
// WAL-2: [A, B] A stackIds = [0, 4], B stackIds = [1, 5]
// WAL-1: [A, B, C, D]
//
// In the case above we need to read one more WAL to be able to consider
// the root procedure A and all children as ready.
// ==============================================================================================
private final WALProcedureMap localProcedureMap = new WALProcedureMap(1024);
private final WALProcedureMap procedureMap = new WALProcedureMap(1024);
/**
* We will use the localProcedureMap to track the active procedures for the current proc wal file,
* and when we finished reading one proc wal file, we will merge he localProcedureMap to the
* procedureMap, which tracks the global active procedures.
* <p/>
* See the comments of {@link WALProcedureMap} for more details.
* <p/>
* After reading all the proc wal files, we will use the procedures in the procedureMap to build a
* {@link WALProcedureTree}, and then give the result to the upper layer. See the comments of
* {@link WALProcedureTree} and the code in {@link #finish()} for more details.
*/
private final WALProcedureMap localProcedureMap = new WALProcedureMap();
private final WALProcedureMap procedureMap = new WALProcedureMap();
private final ProcedureWALFormat.Loader loader;
@ -178,7 +132,7 @@ public class ProcedureWALFormatReader {
localTracker.setMinMaxModifiedProcIds(localProcedureMap.getMinModifiedProcId(),
localProcedureMap.getMaxModifiedProcId());
}
procedureMap.mergeTail(localProcedureMap);
procedureMap.merge(localProcedureMap);
}
if (localTracker.isPartial()) {
localTracker.setPartialFlag(false);
@ -189,18 +143,11 @@ public class ProcedureWALFormatReader {
// notify the loader about the max proc ID
loader.setMaxProcId(maxProcId);
// fetch the procedure ready to run.
ProcedureIterator procIter = procedureMap.fetchReady();
if (procIter != null) {
loader.load(procIter);
}
// remaining procedures have missing link or dependencies
// consider them as corrupted, manual fix is probably required.
procIter = procedureMap.fetchAll();
if (procIter != null) {
loader.handleCorrupted(procIter);
}
// build the procedure execution tree. When building we will verify that whether a procedure is
// valid.
WALProcedureTree tree = WALProcedureTree.build(procedureMap.getProcedures());
loader.load(tree.getValidProcs());
loader.handleCorrupted(tree.getCorruptedProcs());
}
private void setDeletedIfPartial(ProcedureStoreTracker tracker, long procId) {

View File

@ -17,193 +17,50 @@
*/
package org.apache.hadoop.hbase.procedure2.store.wal;
import java.io.IOException;
import org.apache.hadoop.hbase.procedure2.Procedure;
import org.apache.hadoop.hbase.procedure2.ProcedureUtil;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;
import java.util.Collection;
import java.util.Collections;
import java.util.HashMap;
import java.util.Map;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos;
/**
* We keep an in-memory map of the procedures sorted by replay order. (see the details in the
* beginning of {@link ProcedureWALFormatReader}).
*
* <pre>
* procedureMap = | A | | E | | C | | | | | G | | |
* D B
* replayOrderHead = C <-> B <-> E <-> D <-> A <-> G
*
* We also have a lazy grouping by "root procedure", and a list of
* unlinked procedures. If after reading all the WALs we have unlinked
* procedures it means that we had a missing WAL or a corruption.
* rootHead = A <-> D <-> G
* B E
* C
* unlinkFromLinkList = None
* </pre>
* This class is used to track the active procedures when loading procedures from proc wal file.
* <p/>
* We will read proc wal files from new to old, but when reading a proc wal file, we will still read
* from top to bottom, so there are two groups of methods for this class.
* <p/>
* The first group is {@link #add(ProcedureProtos.Procedure)} and {@link #remove(long)}. It is used
* when reading a proc wal file. In these methods, for the same procedure, typically the one comes
* later should win, please see the comment for
* {@link #isIncreasing(ProcedureProtos.Procedure, ProcedureProtos.Procedure)} to see the
* exceptions.
* <p/>
* The second group is {@link #merge(WALProcedureMap)}. We will have a global
* {@link WALProcedureMap} to hold global the active procedures, and a local {@link WALProcedureMap}
* to hold the active procedures for the current proc wal file. And when we finish reading a proc
* wal file, we will merge the local one into the global one, by calling the
* {@link #merge(WALProcedureMap)} method of the global one and pass the local one in. In this
* method, for the same procedure, the one comes earlier will win, as we read the proc wal files
* from new to old(the reverse order).
*/
@InterfaceAudience.Private
class WALProcedureMap {
private static final Logger LOG = LoggerFactory.getLogger(WALProcedureMap.class);
private static class Entry {
// For bucketed linked lists in hash-table.
private Entry hashNext;
// child head
private Entry childHead;
// double-link for rootHead or childHead
private Entry linkNext;
private Entry linkPrev;
// replay double-linked-list
private Entry replayNext;
private Entry replayPrev;
// procedure-infos
private Procedure<?> procedure;
private ProcedureProtos.Procedure proto;
private boolean ready = false;
private final Map<Long, ProcedureProtos.Procedure> procMap = new HashMap<>();
public Entry(Entry hashNext) {
this.hashNext = hashNext;
}
public long getProcId() {
return proto.getProcId();
}
public long getParentId() {
return proto.getParentId();
}
public boolean hasParent() {
return proto.hasParentId();
}
public boolean isReady() {
return ready;
}
public boolean isFinished() {
if (!hasParent()) {
// we only consider 'root' procedures. because for the user 'finished'
// means when everything up to the 'root' is finished.
switch (proto.getState()) {
case ROLLEDBACK:
case SUCCESS:
return true;
default:
break;
}
}
return false;
}
public Procedure<?> convert() throws IOException {
if (procedure == null) {
procedure = ProcedureUtil.convertToProcedure(proto);
}
return procedure;
}
@Override
public String toString() {
final StringBuilder sb = new StringBuilder();
sb.append("Entry(");
sb.append(getProcId());
sb.append(", parentId=");
sb.append(getParentId());
sb.append(", class=");
sb.append(proto.getClassName());
sb.append(")");
return sb.toString();
}
}
private static class EntryIterator implements ProcedureIterator {
private final Entry replayHead;
private Entry current;
public EntryIterator(Entry replayHead) {
this.replayHead = replayHead;
this.current = replayHead;
}
@Override
public void reset() {
this.current = replayHead;
}
@Override
public boolean hasNext() {
return current != null;
}
@Override
public boolean isNextFinished() {
return current != null && current.isFinished();
}
@Override
public void skipNext() {
current = current.replayNext;
}
@Override
public Procedure<?> next() throws IOException {
try {
return current.convert();
} finally {
current = current.replayNext;
}
}
}
// procedure hash table
private Entry[] procedureMap;
// replay-order double-linked-list
private Entry replayOrderHead;
private Entry replayOrderTail;
// root linked-list
private Entry rootHead;
// pending unlinked children (root not present yet)
private Entry childUnlinkedHead;
// Track ProcId range
private long minModifiedProcId = Long.MAX_VALUE;
private long maxModifiedProcId = Long.MIN_VALUE;
public WALProcedureMap(int size) {
procedureMap = new Entry[size];
replayOrderHead = null;
replayOrderTail = null;
rootHead = null;
childUnlinkedHead = null;
}
public void add(ProcedureProtos.Procedure procProto) {
trackProcIds(procProto.getProcId());
Entry entry = addToMap(procProto.getProcId(), procProto.hasParentId());
boolean newEntry = entry.proto == null;
// We have seen procedure WALs where the entries are out of order; see HBASE-18152.
// To compensate, only replace the Entry procedure if for sure this new procedure
// is indeed an entry that came later.
// TODO: Fix the writing of procedure info so it does not violate basic expectation, that WALs
// contain procedure changes goingfrom start to finish in sequence.
if (newEntry || isIncreasing(entry.proto, procProto)) {
entry.proto = procProto;
}
addToReplayList(entry);
if (newEntry) {
if (procProto.hasParentId()) {
childUnlinkedHead = addToLinkList(entry, childUnlinkedHead);
} else {
rootHead = addToLinkList(entry, rootHead);
}
}
private void trackProcId(long procId) {
minModifiedProcId = Math.min(minModifiedProcId, procId);
maxModifiedProcId = Math.max(maxModifiedProcId, procId);
}
/**
@ -225,20 +82,44 @@ class WALProcedureMap {
return increasing;
}
public boolean remove(long procId) {
trackProcIds(procId);
Entry entry = removeFromMap(procId);
if (entry != null) {
unlinkFromReplayList(entry);
unlinkFromLinkList(entry);
return true;
}
return false;
public void add(ProcedureProtos.Procedure proc) {
procMap.compute(proc.getProcId(), (procId, existingProc) -> {
if (existingProc == null || isIncreasing(existingProc, proc)) {
return proc;
} else {
return existingProc;
}
});
trackProcId(proc.getProcId());
}
private void trackProcIds(long procId) {
minModifiedProcId = Math.min(minModifiedProcId, procId);
maxModifiedProcId = Math.max(maxModifiedProcId, procId);
public void remove(long procId) {
procMap.remove(procId);
}
public boolean isEmpty() {
return procMap.isEmpty();
}
public boolean contains(long procId) {
return procMap.containsKey(procId);
}
/**
* Merge the given {@link WALProcedureMap} into this one. The {@link WALProcedureMap} passed in
* will be cleared after merging.
*/
public void merge(WALProcedureMap other) {
other.procMap.forEach(procMap::putIfAbsent);
maxModifiedProcId = Math.max(maxModifiedProcId, other.maxModifiedProcId);
minModifiedProcId = Math.max(minModifiedProcId, other.minModifiedProcId);
other.procMap.clear();
other.maxModifiedProcId = Long.MIN_VALUE;
other.minModifiedProcId = Long.MAX_VALUE;
}
public Collection<ProcedureProtos.Procedure> getProcedures() {
return Collections.unmodifiableCollection(procMap.values());
}
public long getMinModifiedProcId() {
@ -248,360 +129,4 @@ class WALProcedureMap {
public long getMaxModifiedProcId() {
return maxModifiedProcId;
}
public boolean contains(long procId) {
return getProcedure(procId) != null;
}
public boolean isEmpty() {
return replayOrderHead == null;
}
public void clear() {
for (int i = 0; i < procedureMap.length; ++i) {
procedureMap[i] = null;
}
replayOrderHead = null;
replayOrderTail = null;
rootHead = null;
childUnlinkedHead = null;
minModifiedProcId = Long.MAX_VALUE;
maxModifiedProcId = Long.MIN_VALUE;
}
/*
* Merges two WalProcedureMap, the target is the "global" map, the source is the "local" map. -
* The entries in the hashtables are guaranteed to be unique. On replay we don't load procedures
* that already exist in the "global" map (the one we are merging the "local" in to). - The
* replayOrderList of the "local" nao will be appended to the "global" map replay list. - The
* "local" map will be cleared at the end of the operation.
*/
public void mergeTail(WALProcedureMap other) {
for (Entry p = other.replayOrderHead; p != null; p = p.replayNext) {
int slotIndex = getMapSlot(p.getProcId());
p.hashNext = procedureMap[slotIndex];
procedureMap[slotIndex] = p;
}
if (replayOrderHead == null) {
replayOrderHead = other.replayOrderHead;
replayOrderTail = other.replayOrderTail;
rootHead = other.rootHead;
childUnlinkedHead = other.childUnlinkedHead;
} else {
// append replay list
assert replayOrderTail.replayNext == null;
assert other.replayOrderHead.replayPrev == null;
replayOrderTail.replayNext = other.replayOrderHead;
other.replayOrderHead.replayPrev = replayOrderTail;
replayOrderTail = other.replayOrderTail;
// merge rootHead
if (rootHead == null) {
rootHead = other.rootHead;
} else if (other.rootHead != null) {
Entry otherTail = findLinkListTail(other.rootHead);
otherTail.linkNext = rootHead;
rootHead.linkPrev = otherTail;
rootHead = other.rootHead;
}
// merge childUnlinkedHead
if (childUnlinkedHead == null) {
childUnlinkedHead = other.childUnlinkedHead;
} else if (other.childUnlinkedHead != null) {
Entry otherTail = findLinkListTail(other.childUnlinkedHead);
otherTail.linkNext = childUnlinkedHead;
childUnlinkedHead.linkPrev = otherTail;
childUnlinkedHead = other.childUnlinkedHead;
}
}
maxModifiedProcId = Math.max(maxModifiedProcId, other.maxModifiedProcId);
minModifiedProcId = Math.max(minModifiedProcId, other.minModifiedProcId);
other.clear();
}
/**
* Returns an EntryIterator with the list of procedures ready to be added to the executor. A
* Procedure is ready if its children and parent are ready.
*/
public ProcedureIterator fetchReady() {
buildGraph();
Entry readyHead = null;
Entry readyTail = null;
Entry p = replayOrderHead;
while (p != null) {
Entry next = p.replayNext;
if (p.isReady()) {
unlinkFromReplayList(p);
if (readyTail != null) {
readyTail.replayNext = p;
p.replayPrev = readyTail;
} else {
p.replayPrev = null;
readyHead = p;
}
readyTail = p;
p.replayNext = null;
}
p = next;
}
// we need the hash-table lookups for parents, so this must be done
// out of the loop where we check isReadyToRun()
for (p = readyHead; p != null; p = p.replayNext) {
removeFromMap(p.getProcId());
unlinkFromLinkList(p);
}
return readyHead != null ? new EntryIterator(readyHead) : null;
}
/**
* Drain this map and return all procedures in it.
*/
public ProcedureIterator fetchAll() {
Entry head = replayOrderHead;
for (Entry p = head; p != null; p = p.replayNext) {
removeFromMap(p.getProcId());
}
for (int i = 0; i < procedureMap.length; ++i) {
assert procedureMap[i] == null : "map not empty i=" + i;
}
replayOrderHead = null;
replayOrderTail = null;
childUnlinkedHead = null;
rootHead = null;
return head != null ? new EntryIterator(head) : null;
}
private void buildGraph() {
Entry p = childUnlinkedHead;
while (p != null) {
Entry next = p.linkNext;
Entry rootProc = getRootProcedure(p);
if (rootProc != null) {
rootProc.childHead = addToLinkList(p, rootProc.childHead);
}
p = next;
}
for (p = rootHead; p != null; p = p.linkNext) {
checkReadyToRun(p);
}
}
private Entry getRootProcedure(Entry entry) {
while (entry != null && entry.hasParent()) {
entry = getProcedure(entry.getParentId());
}
return entry;
}
/**
* (see the comprehensive explanation in the beginning of {@link ProcedureWALFormatReader}). A
* Procedure is ready when parent and children are ready. "ready" means that we all the
* information that we need in-memory.
* <p/>
* Example-1:<br/>
* We have two WALs, we start reading from the newest (wal-2)
*
* <pre>
* wal-2 | C B |
* wal-1 | A B C |
* </pre>
*
* If C and B don't depend on A (A is not the parent), we can start them before reading wal-1. If
* B is the only one with parent A we can start C. We have to read one more WAL before being able
* to start B.
* <p/>
* How do we know with the only information in B that we are not ready.
* <ul>
* <li>easy case, the parent is missing from the global map</li>
* <li>more complex case we look at the Stack IDs.</li>
* </ul>
* The Stack-IDs are added to the procedure order as an incremental index tracking how many times
* that procedure was executed, which is equivalent to the number of times we wrote the procedure
* to the WAL. <br/>
* In the example above:
*
* <pre>
* wal-2: B has stackId = [1, 2]
* wal-1: B has stackId = [1]
* wal-1: A has stackId = [0]
* </pre>
*
* Since we know that the Stack-IDs are incremental for a Procedure, we notice that there is a gap
* in the stackIds of B, so something was executed before.
* <p/>
* To identify when a Procedure is ready we do the sum of the stackIds of the procedure and the
* parent. if the stackIdSum is equal to the sum of {1..maxStackId} then everything we need is
* available.
* <p/>
* Example-2
*
* <pre>
* wal-2 | A | A stackIds = [0, 2]
* wal-1 | A B | B stackIds = [1]
* </pre>
*
* There is a gap between A stackIds so something was executed in between.
*/
private boolean checkReadyToRun(Entry rootEntry) {
assert !rootEntry.hasParent() : "expected root procedure, got " + rootEntry;
if (rootEntry.isFinished()) {
// If the root procedure is finished, sub-procedures should be gone
if (rootEntry.childHead != null) {
LOG.error("unexpected active children for root-procedure: {}", rootEntry);
for (Entry p = rootEntry.childHead; p != null; p = p.linkNext) {
LOG.error("unexpected active children: {}", p);
}
}
assert rootEntry.childHead == null : "unexpected children on root completion. " + rootEntry;
rootEntry.ready = true;
return true;
}
int stackIdSum = 0;
int maxStackId = 0;
for (int i = 0; i < rootEntry.proto.getStackIdCount(); ++i) {
int stackId = 1 + rootEntry.proto.getStackId(i);
maxStackId = Math.max(maxStackId, stackId);
stackIdSum += stackId;
LOG.trace("stackId={} stackIdSum={} maxStackid={} {}", stackId, stackIdSum, maxStackId,
rootEntry);
}
for (Entry p = rootEntry.childHead; p != null; p = p.linkNext) {
for (int i = 0; i < p.proto.getStackIdCount(); ++i) {
int stackId = 1 + p.proto.getStackId(i);
maxStackId = Math.max(maxStackId, stackId);
stackIdSum += stackId;
LOG.trace("stackId={} stackIdSum={} maxStackid={} {}", stackId, stackIdSum, maxStackId, p);
}
}
// The cmpStackIdSum is this formula for finding the sum of a series of numbers:
// http://www.wikihow.com/Sum-the-Integers-from-1-to-N#/Image:Sum-the-Integers-from-1-to-N-Step-2-Version-3.jpg
final int cmpStackIdSum = (maxStackId * (maxStackId + 1) / 2);
if (cmpStackIdSum == stackIdSum) {
rootEntry.ready = true;
for (Entry p = rootEntry.childHead; p != null; p = p.linkNext) {
p.ready = true;
}
return true;
}
return false;
}
private void unlinkFromReplayList(Entry entry) {
if (replayOrderHead == entry) {
replayOrderHead = entry.replayNext;
}
if (replayOrderTail == entry) {
replayOrderTail = entry.replayPrev;
}
if (entry.replayPrev != null) {
entry.replayPrev.replayNext = entry.replayNext;
}
if (entry.replayNext != null) {
entry.replayNext.replayPrev = entry.replayPrev;
}
}
private void addToReplayList(final Entry entry) {
unlinkFromReplayList(entry);
entry.replayNext = replayOrderHead;
entry.replayPrev = null;
if (replayOrderHead != null) {
replayOrderHead.replayPrev = entry;
} else {
replayOrderTail = entry;
}
replayOrderHead = entry;
}
private void unlinkFromLinkList(Entry entry) {
if (entry == rootHead) {
rootHead = entry.linkNext;
} else if (entry == childUnlinkedHead) {
childUnlinkedHead = entry.linkNext;
}
if (entry.linkPrev != null) {
entry.linkPrev.linkNext = entry.linkNext;
}
if (entry.linkNext != null) {
entry.linkNext.linkPrev = entry.linkPrev;
}
}
private Entry addToLinkList(Entry entry, Entry linkHead) {
unlinkFromLinkList(entry);
entry.linkNext = linkHead;
entry.linkPrev = null;
if (linkHead != null) {
linkHead.linkPrev = entry;
}
return entry;
}
private Entry findLinkListTail(Entry linkHead) {
Entry tail = linkHead;
while (tail.linkNext != null) {
tail = tail.linkNext;
}
return tail;
}
private Entry addToMap(long procId, boolean hasParent) {
int slotIndex = getMapSlot(procId);
Entry entry = getProcedure(slotIndex, procId);
if (entry != null) {
return entry;
}
entry = new Entry(procedureMap[slotIndex]);
procedureMap[slotIndex] = entry;
return entry;
}
private Entry removeFromMap(final long procId) {
int slotIndex = getMapSlot(procId);
Entry prev = null;
Entry entry = procedureMap[slotIndex];
while (entry != null) {
if (procId == entry.getProcId()) {
if (prev != null) {
prev.hashNext = entry.hashNext;
} else {
procedureMap[slotIndex] = entry.hashNext;
}
entry.hashNext = null;
return entry;
}
prev = entry;
entry = entry.hashNext;
}
return null;
}
private Entry getProcedure(long procId) {
return getProcedure(getMapSlot(procId), procId);
}
private Entry getProcedure(int slotIndex, long procId) {
Entry entry = procedureMap[slotIndex];
while (entry != null) {
if (procId == entry.getProcId()) {
return entry;
}
entry = entry.hashNext;
}
return null;
}
private int getMapSlot(long procId) {
return (int) (Procedure.getProcIdHashCode(procId) % procedureMap.length);
}
}
}

View File

@ -0,0 +1,299 @@
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.procedure2.store.wal;
import java.io.IOException;
import java.util.ArrayList;
import java.util.Collection;
import java.util.Collections;
import java.util.Comparator;
import java.util.HashMap;
import java.util.Iterator;
import java.util.List;
import java.util.Map;
import java.util.NoSuchElementException;
import org.apache.commons.lang3.mutable.MutableInt;
import org.apache.hadoop.hbase.procedure2.Procedure;
import org.apache.hadoop.hbase.procedure2.ProcedureUtil;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;
import org.apache.yetus.audience.InterfaceAudience;
import org.slf4j.Logger;
import org.slf4j.LoggerFactory;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos;
/**
* Used to build the tree for procedures.
* <p/>
* We will group the procedures with the root procedure, and then validate each group. For each
* group of procedures(with the same root procedure), we will collect all the stack ids, if the max
* stack id is n, then all the stack ids should be from 0 to n, non-repetition and non-omission. If
* not, we will consider all the procedures in this group as corrupted. Please see the code in
* {@link #checkReady(Entry, Map)} method.
* <p/>
* For the procedures not in any group, i.e, can not find the root procedure for these procedures,
* we will also consider them as corrupted. Please see the code in {@link #checkOrphan(Map)} method.
*/
@InterfaceAudience.Private
public final class WALProcedureTree {
private static final Logger LOG = LoggerFactory.getLogger(WALProcedureTree.class);
private static final class Entry {
private final ProcedureProtos.Procedure proc;
private final List<Entry> subProcs = new ArrayList<>();
public Entry(ProcedureProtos.Procedure proc) {
this.proc = proc;
}
@Override
public String toString() {
StringBuilder sb = new StringBuilder();
sb.append("Procedure(pid=");
sb.append(proc.getProcId());
sb.append(", ppid=");
sb.append(proc.hasParentId() ? proc.getParentId() : Procedure.NO_PROC_ID);
sb.append(", class=");
sb.append(proc.getClassName());
sb.append(")");
return sb.toString();
}
}
// when loading we will iterator the procedures twice, so use this class to cache the deserialized
// result to prevent deserializing multiple times.
private static final class ProtoAndProc {
private final ProcedureProtos.Procedure proto;
private Procedure<?> proc;
public ProtoAndProc(ProcedureProtos.Procedure proto) {
this.proto = proto;
}
public Procedure<?> getProc() throws IOException {
if (proc == null) {
proc = ProcedureUtil.convertToProcedure(proto);
}
return proc;
}
}
private final List<ProtoAndProc> validProcs = new ArrayList<>();
private final List<ProtoAndProc> corruptedProcs = new ArrayList<>();
private static boolean isFinished(ProcedureProtos.Procedure proc) {
if (!proc.hasParentId()) {
switch (proc.getState()) {
case ROLLEDBACK:
case SUCCESS:
return true;
default:
break;
}
}
return false;
}
private WALProcedureTree(Map<Long, Entry> procMap) {
List<Entry> rootEntries = buildTree(procMap);
for (Entry rootEntry : rootEntries) {
checkReady(rootEntry, procMap);
}
checkOrphan(procMap);
Comparator<ProtoAndProc> cmp =
(p1, p2) -> Long.compare(p1.proto.getProcId(), p2.proto.getProcId());
Collections.sort(validProcs, cmp);
Collections.sort(corruptedProcs, cmp);
}
private List<Entry> buildTree(Map<Long, Entry> procMap) {
List<Entry> rootEntries = new ArrayList<>();
procMap.values().forEach(entry -> {
if (!entry.proc.hasParentId()) {
rootEntries.add(entry);
} else {
Entry parentEntry = procMap.get(entry.proc.getParentId());
// For a valid procedure this should not be null. We will log the error later if it is null,
// as it will not be referenced by any root procedures.
if (parentEntry != null) {
parentEntry.subProcs.add(entry);
}
}
});
return rootEntries;
}
private void collectStackId(Entry entry, Map<Integer, List<Entry>> stackId2Proc,
MutableInt maxStackId) {
for (int i = 0, n = entry.proc.getStackIdCount(); i < n; i++) {
int stackId = entry.proc.getStackId(i);
if (stackId > maxStackId.intValue()) {
maxStackId.setValue(stackId);
}
stackId2Proc.computeIfAbsent(stackId, k -> new ArrayList<>()).add(entry);
}
entry.subProcs.forEach(e -> collectStackId(e, stackId2Proc, maxStackId));
}
private void addAllToCorruptedAndRemoveFromProcMap(Entry entry,
Map<Long, Entry> remainingProcMap) {
corruptedProcs.add(new ProtoAndProc(entry.proc));
remainingProcMap.remove(entry.proc.getProcId());
for (Entry e : entry.subProcs) {
addAllToCorruptedAndRemoveFromProcMap(e, remainingProcMap);
}
}
private void addAllToValidAndRemoveFromProcMap(Entry entry, Map<Long, Entry> remainingProcMap) {
validProcs.add(new ProtoAndProc(entry.proc));
remainingProcMap.remove(entry.proc.getProcId());
for (Entry e : entry.subProcs) {
addAllToValidAndRemoveFromProcMap(e, remainingProcMap);
}
}
// In this method first we will check whether the given root procedure and all its sub procedures
// are valid, through the procedure stack. And we will also remove all these procedures from the
// remainingProcMap, so at last, if there are still procedures in the map, we know that there are
// orphan procedures.
private void checkReady(Entry rootEntry, Map<Long, Entry> remainingProcMap) {
if (isFinished(rootEntry.proc)) {
if (!rootEntry.subProcs.isEmpty()) {
LOG.error("unexpected active children for root-procedure: {}", rootEntry);
rootEntry.subProcs.forEach(e -> LOG.error("unexpected active children: {}", e));
addAllToCorruptedAndRemoveFromProcMap(rootEntry, remainingProcMap);
} else {
addAllToValidAndRemoveFromProcMap(rootEntry, remainingProcMap);
}
return;
}
Map<Integer, List<Entry>> stackId2Proc = new HashMap<>();
MutableInt maxStackId = new MutableInt(Integer.MIN_VALUE);
collectStackId(rootEntry, stackId2Proc, maxStackId);
// the stack ids should start from 0 and increase by one every time
boolean valid = true;
for (int i = 0; i <= maxStackId.intValue(); i++) {
List<Entry> entries = stackId2Proc.get(i);
if (entries == null) {
LOG.error("Missing stack id {}, max stack id is {}, root procedure is {}", i, maxStackId,
rootEntry);
valid = false;
} else if (entries.size() > 1) {
LOG.error("Multiple procedures {} have the same stack id {}, max stack id is {}," +
" root procedure is {}", entries, i, maxStackId, rootEntry);
valid = false;
}
}
if (valid) {
addAllToValidAndRemoveFromProcMap(rootEntry, remainingProcMap);
} else {
addAllToCorruptedAndRemoveFromProcMap(rootEntry, remainingProcMap);
}
}
private void checkOrphan(Map<Long, Entry> procMap) {
procMap.values().forEach(entry -> {
LOG.error("Orphan procedure: {}", entry);
corruptedProcs.add(new ProtoAndProc(entry.proc));
});
}
private static final class Iter implements ProcedureIterator {
private final List<ProtoAndProc> procs;
private Iterator<ProtoAndProc> iter;
private ProtoAndProc current;
public Iter(List<ProtoAndProc> procs) {
this.procs = procs;
reset();
}
@Override
public void reset() {
iter = procs.iterator();
if (iter.hasNext()) {
current = iter.next();
} else {
current = null;
}
}
@Override
public boolean hasNext() {
return current != null;
}
private void checkNext() {
if (!hasNext()) {
throw new NoSuchElementException();
}
}
@Override
public boolean isNextFinished() {
checkNext();
return isFinished(current.proto);
}
private void moveToNext() {
if (iter.hasNext()) {
current = iter.next();
} else {
current = null;
}
}
@Override
public void skipNext() {
checkNext();
moveToNext();
}
@Override
public Procedure<?> next() throws IOException {
checkNext();
Procedure<?> proc = current.getProc();
moveToNext();
return proc;
}
}
public ProcedureIterator getValidProcs() {
return new Iter(validProcs);
}
public ProcedureIterator getCorruptedProcs() {
return new Iter(corruptedProcs);
}
public static WALProcedureTree build(Collection<ProcedureProtos.Procedure> procedures) {
Map<Long, Entry> procMap = new HashMap<>();
for (ProcedureProtos.Procedure proc : procedures) {
procMap.put(proc.getProcId(), new Entry(proc));
}
return new WALProcedureTree(procMap);
}
}

View File

@ -18,9 +18,7 @@
package org.apache.hadoop.hbase.procedure2.store.wal;
import static org.junit.Assert.assertEquals;
import static org.junit.Assert.assertFalse;
import static org.junit.Assert.assertTrue;
import static org.junit.Assert.fail;
import java.io.IOException;
import java.util.Random;

View File

@ -20,7 +20,6 @@ package org.apache.hadoop.hbase.procedure2.store.wal;
import static org.junit.Assert.assertEquals;
import static org.junit.Assert.assertFalse;
import static org.junit.Assert.assertTrue;
import static org.junit.Assert.fail;
import java.io.FileNotFoundException;
import java.io.IOException;
@ -44,7 +43,6 @@ import org.apache.hadoop.hbase.procedure2.ProcedureTestingUtility.LoadCounter;
import org.apache.hadoop.hbase.procedure2.ProcedureTestingUtility.TestProcedure;
import org.apache.hadoop.hbase.procedure2.SequentialProcedure;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStoreTracker;
import org.apache.hadoop.hbase.testclassification.MasterTests;
import org.apache.hadoop.hbase.testclassification.SmallTests;
@ -72,7 +70,6 @@ public class TestWALProcedureStore {
private static final Logger LOG = LoggerFactory.getLogger(TestWALProcedureStore.class);
private static final int PROCEDURE_STORE_SLOTS = 1;
private static final Procedure NULL_PROC = null;
private WALProcedureStore procStore;
@ -153,7 +150,7 @@ public class TestWALProcedureStore {
@Test
public void testWalCleanerSequentialClean() throws Exception {
final Procedure[] procs = new Procedure[5];
final Procedure<?>[] procs = new Procedure[5];
ArrayList<ProcedureWALFile> logs = null;
// Insert procedures and roll wal after every insert.
@ -182,7 +179,7 @@ public class TestWALProcedureStore {
// they are in the starting of the list.
@Test
public void testWalCleanerNoHoles() throws Exception {
final Procedure[] procs = new Procedure[5];
final Procedure<?>[] procs = new Procedure[5];
ArrayList<ProcedureWALFile> logs = null;
// Insert procedures and roll wal after every insert.
for (int i = 0; i < procs.length; i++) {
@ -242,7 +239,7 @@ public class TestWALProcedureStore {
@Test
public void testWalCleanerWithEmptyRolls() throws Exception {
final Procedure[] procs = new Procedure[3];
final Procedure<?>[] procs = new Procedure[3];
for (int i = 0; i < procs.length; ++i) {
procs[i] = new TestSequentialProcedure();
procStore.insert(procs[i], null);
@ -284,12 +281,12 @@ public class TestWALProcedureStore {
Set<Long> procIds = new HashSet<>();
// Insert something in the log
Procedure proc1 = new TestSequentialProcedure();
Procedure<?> proc1 = new TestSequentialProcedure();
procIds.add(proc1.getProcId());
procStore.insert(proc1, null);
Procedure proc2 = new TestSequentialProcedure();
Procedure[] child2 = new Procedure[2];
Procedure<?> proc2 = new TestSequentialProcedure();
Procedure<?>[] child2 = new Procedure[2];
child2[0] = new TestSequentialProcedure();
child2[1] = new TestSequentialProcedure();
@ -323,11 +320,11 @@ public class TestWALProcedureStore {
@Test
public void testNoTrailerDoubleRestart() throws Exception {
// log-0001: proc 0, 1 and 2 are inserted
Procedure proc0 = new TestSequentialProcedure();
Procedure<?> proc0 = new TestSequentialProcedure();
procStore.insert(proc0, null);
Procedure proc1 = new TestSequentialProcedure();
Procedure<?> proc1 = new TestSequentialProcedure();
procStore.insert(proc1, null);
Procedure proc2 = new TestSequentialProcedure();
Procedure<?> proc2 = new TestSequentialProcedure();
procStore.insert(proc2, null);
procStore.rollWriterForTesting();
@ -420,7 +417,7 @@ public class TestWALProcedureStore {
}
private static void assertUpdated(final ProcedureStoreTracker tracker,
final Procedure[] procs, final int[] updatedProcs, final int[] nonUpdatedProcs) {
final Procedure<?>[] procs, final int[] updatedProcs, final int[] nonUpdatedProcs) {
for (int index : updatedProcs) {
long procId = procs[index].getProcId();
assertTrue("Procedure id : " + procId, tracker.isModified(procId));
@ -432,7 +429,7 @@ public class TestWALProcedureStore {
}
private static void assertDeleted(final ProcedureStoreTracker tracker,
final Procedure[] procs, final int[] deletedProcs, final int[] nonDeletedProcs) {
final Procedure<?>[] procs, final int[] deletedProcs, final int[] nonDeletedProcs) {
for (int index : deletedProcs) {
long procId = procs[index].getProcId();
assertEquals("Procedure id : " + procId,
@ -447,7 +444,7 @@ public class TestWALProcedureStore {
@Test
public void testCorruptedTrailersRebuild() throws Exception {
final Procedure[] procs = new Procedure[6];
final Procedure<?>[] procs = new Procedure[6];
for (int i = 0; i < procs.length; ++i) {
procs[i] = new TestSequentialProcedure();
}
@ -575,127 +572,20 @@ public class TestWALProcedureStore {
storeRestart(loader);
assertEquals(0, loader.getLoadedCount());
assertEquals(rootProcs.length, loader.getCorruptedCount());
for (Procedure proc: loader.getCorrupted()) {
for (Procedure<?> proc : loader.getCorrupted()) {
assertTrue(proc.toString(), proc.getParentProcId() <= rootProcs.length);
assertTrue(proc.toString(),
proc.getProcId() > rootProcs.length &&
proc.getProcId() <= (rootProcs.length * 2));
proc.getProcId() > rootProcs.length && proc.getProcId() <= (rootProcs.length * 2));
}
}
@Test
public void testWalReplayOrder_AB_A() throws Exception {
/*
* | A B | -> | A |
*/
TestProcedure a = new TestProcedure(1, 0);
TestProcedure b = new TestProcedure(2, 1);
procStore.insert(a, null);
a.addStackId(0);
procStore.update(a);
procStore.insert(a, new Procedure[] { b });
b.addStackId(1);
procStore.update(b);
procStore.rollWriterForTesting();
a.addStackId(2);
procStore.update(a);
storeRestart(new ProcedureStore.ProcedureLoader() {
@Override
public void setMaxProcId(long maxProcId) {
assertEquals(2, maxProcId);
}
@Override
public void load(ProcedureIterator procIter) throws IOException {
assertTrue(procIter.hasNext());
assertEquals(1, procIter.next().getProcId());
assertTrue(procIter.hasNext());
assertEquals(2, procIter.next().getProcId());
assertFalse(procIter.hasNext());
}
@Override
public void handleCorrupted(ProcedureIterator procIter) throws IOException {
assertFalse(procIter.hasNext());
}
});
}
@Test
public void testWalReplayOrder_ABC_BAD() throws Exception {
/*
* | A B C | -> | B A D |
*/
TestProcedure a = new TestProcedure(1, 0);
TestProcedure b = new TestProcedure(2, 1);
TestProcedure c = new TestProcedure(3, 2);
TestProcedure d = new TestProcedure(4, 0);
procStore.insert(a, null);
a.addStackId(0);
procStore.update(a);
procStore.insert(a, new Procedure[] { b });
b.addStackId(1);
procStore.update(b);
procStore.insert(b, new Procedure[] { c });
b.addStackId(2);
procStore.update(b);
procStore.rollWriterForTesting();
b.addStackId(3);
procStore.update(b);
a.addStackId(4);
procStore.update(a);
procStore.insert(d, null);
d.addStackId(0);
procStore.update(d);
storeRestart(new ProcedureStore.ProcedureLoader() {
@Override
public void setMaxProcId(long maxProcId) {
assertEquals(4, maxProcId);
}
@Override
public void load(ProcedureIterator procIter) throws IOException {
assertTrue(procIter.hasNext());
assertEquals(4, procIter.next().getProcId());
// TODO: This will be multiple call once we do fast-start
//assertFalse(procIter.hasNext());
assertTrue(procIter.hasNext());
assertEquals(1, procIter.next().getProcId());
assertTrue(procIter.hasNext());
assertEquals(2, procIter.next().getProcId());
assertTrue(procIter.hasNext());
assertEquals(3, procIter.next().getProcId());
assertFalse(procIter.hasNext());
}
@Override
public void handleCorrupted(ProcedureIterator procIter) throws IOException {
assertFalse(procIter.hasNext());
}
});
}
@Test
public void testRollAndRemove() throws IOException {
// Insert something in the log
Procedure proc1 = new TestSequentialProcedure();
Procedure<?> proc1 = new TestSequentialProcedure();
procStore.insert(proc1, null);
Procedure proc2 = new TestSequentialProcedure();
Procedure<?> proc2 = new TestSequentialProcedure();
procStore.insert(proc2, null);
// roll the log, now we have 2
@ -942,17 +832,6 @@ public class TestWALProcedureStore {
assertEquals(0, loader.getCorruptedCount());
}
private void assertEmptyLogDir() {
try {
FileStatus[] status = fs.listStatus(logDir);
assertTrue("expected empty state-log dir", status == null || status.length == 0);
} catch (FileNotFoundException e) {
fail("expected the state-log dir to be present: " + logDir);
} catch (IOException e) {
fail("got en exception on state-log dir list: " + e.getMessage());
}
}
public static class TestSequentialProcedure extends SequentialProcedure<Void> {
private static long seqid = 0;
@ -961,13 +840,18 @@ public class TestWALProcedureStore {
}
@Override
protected Procedure[] execute(Void env) { return null; }
protected Procedure<Void>[] execute(Void env) {
return null;
}
@Override
protected void rollback(Void env) { }
protected void rollback(Void env) {
}
@Override
protected boolean abort(Void env) { return false; }
protected boolean abort(Void env) {
return false;
}
@Override
protected void serializeStateData(ProcedureStateSerializer serializer)

View File

@ -0,0 +1,173 @@
/**
* Licensed to the Apache Software Foundation (ASF) under one
* or more contributor license agreements. See the NOTICE file
* distributed with this work for additional information
* regarding copyright ownership. The ASF licenses this file
* to you under the Apache License, Version 2.0 (the
* "License"); you may not use this file except in compliance
* with the License. You may obtain a copy of the License at
*
* http://www.apache.org/licenses/LICENSE-2.0
*
* Unless required by applicable law or agreed to in writing, software
* distributed under the License is distributed on an "AS IS" BASIS,
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
* See the License for the specific language governing permissions and
* limitations under the License.
*/
package org.apache.hadoop.hbase.procedure2.store.wal;
import static org.junit.Assert.assertEquals;
import java.io.IOException;
import java.io.UncheckedIOException;
import java.util.ArrayList;
import java.util.Arrays;
import java.util.List;
import java.util.stream.Collectors;
import org.apache.hadoop.hbase.HBaseClassTestRule;
import org.apache.hadoop.hbase.procedure2.Procedure;
import org.apache.hadoop.hbase.procedure2.ProcedureStateSerializer;
import org.apache.hadoop.hbase.procedure2.ProcedureSuspendedException;
import org.apache.hadoop.hbase.procedure2.ProcedureUtil;
import org.apache.hadoop.hbase.procedure2.ProcedureYieldException;
import org.apache.hadoop.hbase.procedure2.store.ProcedureStore.ProcedureIterator;
import org.apache.hadoop.hbase.testclassification.MasterTests;
import org.apache.hadoop.hbase.testclassification.SmallTests;
import org.junit.ClassRule;
import org.junit.Test;
import org.junit.experimental.categories.Category;
import org.apache.hadoop.hbase.shaded.protobuf.generated.ProcedureProtos;
@Category({ MasterTests.class, SmallTests.class })
public class TestWALProcedureTree {
@ClassRule
public static final HBaseClassTestRule CLASS_RULE =
HBaseClassTestRule.forClass(TestWALProcedureTree.class);
public static final class TestProcedure extends Procedure<Void> {
@Override
public void setProcId(long procId) {
super.setProcId(procId);
}
@Override
public void setParentProcId(long parentProcId) {
super.setParentProcId(parentProcId);
}
@Override
public synchronized void addStackIndex(int index) {
super.addStackIndex(index);
}
@Override
protected Procedure<Void>[] execute(Void env)
throws ProcedureYieldException, ProcedureSuspendedException, InterruptedException {
return null;
}
@Override
protected void rollback(Void env) throws IOException, InterruptedException {
}
@Override
protected boolean abort(Void env) {
return false;
}
@Override
protected void serializeStateData(ProcedureStateSerializer serializer) throws IOException {
}
@Override
protected void deserializeStateData(ProcedureStateSerializer serializer) throws IOException {
}
}
private TestProcedure createProc(long procId, long parentProcId) {
TestProcedure proc = new TestProcedure();
proc.setProcId(procId);
if (parentProcId != Procedure.NO_PROC_ID) {
proc.setParentProcId(parentProcId);
}
return proc;
}
private List<ProcedureProtos.Procedure> toProtos(TestProcedure... procs) {
return Arrays.stream(procs).map(p -> {
try {
return ProcedureUtil.convertToProtoProcedure(p);
} catch (IOException e) {
throw new UncheckedIOException(e);
}
}).collect(Collectors.toList());
}
private List<TestProcedure> getProcs(ProcedureIterator iter) throws IOException {
List<TestProcedure> procs = new ArrayList<>();
while (iter.hasNext()) {
procs.add((TestProcedure) iter.next());
}
return procs;
}
@Test
public void testMissingStackId() throws IOException {
TestProcedure proc0 = createProc(1, Procedure.NO_PROC_ID);
proc0.addStackIndex(0);
TestProcedure proc1 = createProc(2, 1);
proc1.addStackIndex(1);
TestProcedure proc2 = createProc(3, 2);
proc2.addStackIndex(3);
WALProcedureTree tree = WALProcedureTree.build(toProtos(proc0, proc1, proc2));
List<TestProcedure> validProcs = getProcs(tree.getValidProcs());
assertEquals(0, validProcs.size());
List<TestProcedure> corruptedProcs = getProcs(tree.getCorruptedProcs());
assertEquals(3, corruptedProcs.size());
assertEquals(1, corruptedProcs.get(0).getProcId());
assertEquals(2, corruptedProcs.get(1).getProcId());
assertEquals(3, corruptedProcs.get(2).getProcId());
}
@Test
public void testDuplicatedStackId() throws IOException {
TestProcedure proc0 = createProc(1, Procedure.NO_PROC_ID);
proc0.addStackIndex(0);
TestProcedure proc1 = createProc(2, 1);
proc1.addStackIndex(1);
TestProcedure proc2 = createProc(3, 2);
proc2.addStackIndex(1);
WALProcedureTree tree = WALProcedureTree.build(toProtos(proc0, proc1, proc2));
List<TestProcedure> validProcs = getProcs(tree.getValidProcs());
assertEquals(0, validProcs.size());
List<TestProcedure> corruptedProcs = getProcs(tree.getCorruptedProcs());
assertEquals(3, corruptedProcs.size());
assertEquals(1, corruptedProcs.get(0).getProcId());
assertEquals(2, corruptedProcs.get(1).getProcId());
assertEquals(3, corruptedProcs.get(2).getProcId());
}
@Test
public void testOrphan() throws IOException {
TestProcedure proc0 = createProc(1, Procedure.NO_PROC_ID);
proc0.addStackIndex(0);
TestProcedure proc1 = createProc(2, 1);
proc1.addStackIndex(1);
TestProcedure proc2 = createProc(3, Procedure.NO_PROC_ID);
proc2.addStackIndex(0);
TestProcedure proc3 = createProc(5, 4);
proc3.addStackIndex(1);
WALProcedureTree tree = WALProcedureTree.build(toProtos(proc0, proc1, proc2, proc3));
List<TestProcedure> validProcs = getProcs(tree.getValidProcs());
assertEquals(3, validProcs.size());
List<TestProcedure> corruptedProcs = getProcs(tree.getCorruptedProcs());
assertEquals(1, corruptedProcs.size());
assertEquals(5, corruptedProcs.get(0).getProcId());
assertEquals(4, corruptedProcs.get(0).getParentProcId());
}
}

View File

@ -171,6 +171,7 @@ import org.apache.hadoop.hbase.shaded.protobuf.ProtobufUtil;
* avoiding port contention if another local HBase instance is already running).
* <p>To preserve test data directories, pass the system property "hbase.testing.preserve.testdir"
* setting it to true.
* Trigger pre commit.
*/
@InterfaceAudience.Public
@SuppressWarnings("deprecation")